Professional Documents
Culture Documents
1 Introduction
The critical uncertainty in the numerical simulation of tur- formance of the k—t model, and to evaluate the improvement
bulent shear flows is the estimation of the Reynolds stresses. of the performance obtained by the Hanjalic and Launder [6]
Models in current use may be as simple as explicit algebraic modification2, for two flows. Both cases were chosen because
relations that connect the stresses to the mean flow variables, of their relevance to airfoil flows. The first case, Simpson's
or as complex as a set of differential transport equations with problem [7, 8, 9, 10], was flow along a wind tunnel wall with
one equation for each stress. The k — t model of Launder and an imposed pressure distribution that approximated the
Spalding [1] has emerged as a standard because it has been pressure on an airfoil. The second case was the flow over a
quite successful for predicting a wide variety of flows, yet is supercritical airfoil measured by Nakayama [11]. Simpson's
reasonably simple. flow experiences a strong adverse pressure gradient but no
Some authors have, however, reported serious deficiencies boundary-induced streamline curvature (internal streamline
in the A: —e model for particular flows. For flows with strong curvature does result from flow separation). Nakayama's flow
streamline curvature, Leschziner and Rodi [2] found that the is subject to both strong adverse pressure gradients and strong
measured anisotropy of the normal stresses and the magnitude streamline curvature.
of the shear stresses are not well predicted by the k—t model. To choose an appropriate turbulence model for a computa-
For flows in a strong adverse pressure gradient, Rodi and tional fluid dynamics code, the analyst must know the
Scheuerer [3] found that the predicted dissipation is too small, strengths and weaknesses of all candidate models. This
with the result that turbulent stresses are too large. knowledge is gained through comparisons of predictions with
Some attempts have been made to modify the k — t model data for a variety of flows. The aim of this paper is to add two
and improve its performance for such flows. Leschziner and case studies to the literature that will assist potential users to
Rodi [2], and Pourahmadi and Humphrey [4] suggested mak- understand more completely the capabilities of the standard
ing the C^ "constant" in the model dependent on the flow k — t model, and the potential benefits of implementing the
curvature. This improved the predictions for jet flows, but Hanjalic and Launder [6] modification.
created convergence difficulties [5] for an internal flow. In
their study of flows with strong adverse pressure gradients, 2 Formulation and Solution Method
Hanjalic and Launder [6] found that a higher production of
dissipation was required to correctly predict the turbulence 2.1 Equations of Motion. For a turbulent, incompressi-
levels for cases where the mean flow is irrotational. This led to ble, two-dimensional flow, the time-averaged differential
a proposal [6, 3] for increasing the coefficient that multiplies equations of motion, written for a general orthogonal coor-
the production of turbulent kinetic energy by normal stresses dinate system, are [12]:
in the e equation. Conservation of mass
The present study was undertaken to examine the per-
{ph2U)+—(PhiV) =0 (1)
h,h' 7• [ • Idx dx-,
Current Address: Scientific Research Associates, Glastonbury, Conn.
Contributed by the Fluids Engineering Division for publication in the JOUR-
NAL OF FLUIDS ENGINEERING. Manuscript received by the Fluids Engineering The model has been generalized in the present paper to make it coordinate
Division December 12, 1988. system independent.
Nomenclature
1 dU dh. where K and E are the log-law constants equal to 0.41 and 9.0
(uu — vv) = — • (for a smooth wall), respectively [1].
P h{ dxx h1h2 dx2
Expressions for the shear stress at the wall are obtained
which follows from the turbulent viscosity definition of from equations (20) and (21), assuming that for the region
stresses. For a fully developed flow between parallel plates, very near the wall the pressure gradient has no influence on the
this equation yields (uu — vv) =0, which is contrary to flow [16] and the flow is locally one-dimensional. The expres-
measurements. Therefore, the present study has followed sions are
Hanjalic and Launder by using equation (16). for^<_vL
To apply this model to a flow not aligned with the xt direc-
:\MklP/2K
tion (assumed in equation (16)), the local s — n coordinate (22)
system in Fig. 1 is introduced in a manner similar to that pro- tanh(K,.Vp)
posed by Leschziner and Rodi [2]. The expression for Pk in for yp~>y\ +
this coordinate system is
pUpC^k1?
Pk = 4n,(esn)2-0.33pkess (17) (23)
where ess and es„ are the normal and shear strains respectively. The subscript P refers to the node in the flow that lies adjacent
These are defined in Appendix A. to the wall while/1 + is the value of y+ at which equations (22)
Equation (17) is used to calculate the production of tur- and (23) give the same value of r„.
bulence kinetic energy in the ^-equation (equation (11)) as well
as in the e-equation (equation (12)) where P e is expressed as Wall Boundary Condition for k and e. At the wall,
dk/dy = 0. In addition, assuming that the dominant term in the
Pc =r[CelPk - ( C d - C€l)0.33PkeJ (18) near wall production is the shear stress, the following can be
The derivation of equations (17) and (18) is given in Appendix derived from equations (20) and (21)
A. for^+<^2 +
2.3 Boundary Conditions. This section deals with the n P2U\ tanh2(Kl>< + )
(24)
application of the boundary conditions for flow along a tur- \x. cosb2(Kiy + )
bulent wall. fory+>y2*
Wall Boundary Conditions on Velocity. The velocity nor- l
Ut 1
mal to a solid wall is set to zero. The boundary condition for (25)
Ky +
the tangential velocity is specified by a shear stress based on a /*
wall function approach described below. y2 + is the value of y+ at which equations (24) and (25) give
In the derivation of the wall function, y+ is taken as the the same value of Pk. The production of k for a given node
non-dimensional distance normal to the wall that lies immediately adjacent to the wall is obtained by in-
pyUT tegrating equations (22) and (23) over the y+ region occupied
(19) by the node.
r Based on the assumption that the flow is in local
If there is an equilibrium layer near the surface, in which the equilibrium in the near wall region (pe = Pk), the variation of e
shear stress is nearly constant and the production of k is in with distance normal to the wall is simply expressed as
balance with its dissipation, then f o r 5 < j + <yl +
111 lll'.cl'.l.lljL' (
I J
Fig. 2 Wind tunnel geometry and fluid properties for the turbulent
separating boundary layer flow problem
0.6
0.4
0.2
0.0 <»*
0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 0.8 1.0
X = 3.996TI
0.0
0.000 0.006 0.000 0.006 0.000 0.006 0.000 0.006
Following the recommendation of Cebeci et al. [21], an inlet model (SKE), and with the modified model (MKE), in which
dissipation rate of e = k3/2/(0.005H) was prescribed, were His Ce3 = 5.50.
the wind tunnel height at the inlet. The wind tunnel walls were
taken as hydraulically smooth. Separation on the upper sur- Presentation and Discussion of the Results. The SKE
face was avoided in the computer model by withdrawing mass model failed to predict separation on the lower surface. The
through the wall between B and C at a uniform rate which MKE model predicted separation at *~3.4m, which lies
matched the measured total withdrawal. This produced within the range of observed separation. To determine the sen-
velocities normal to the wall of about 0.7 percent of the inlet sitivity of the gross features of the flow to Ce3, solutions were
velocity. obtained for several different values of Cei. Flow separation
Numerical solutions were obtained with the standard k — e was found for Ce3>2.50. As Ce3 was increased above this
Experimental x = 3.896m: o ir
60x35 grid : 1
1.0
60x21 grid : IP
95x21 grid : I
0.8 / -
0 J
O y//
TS 0.6 /v -
° /y
0.4
0
/? _
o fl
0.2 - 0 / - 0.0 4.0 8.0 0.0 4.0 B.O 0.0 4.0 8.0 0.0 4.0 8.0
0o / (BU/t/j;) x 1000
/
%JL\ I i i i i Fig. 10(a) uu Profiles for supercritical airfoil at 4 deg incidence.
0.0 1 1
C y
value, the separation bubble increased. For C t3 >5.50, con- 0.0 4.0 8.0 0.0 4.0 B.O 0.0 4.0 8.0 0.0 4.0 8.0
verged solutions could not be obtained. The streamline for the (un/ul) x 1000
MKE model (Ce3 =5.50) are shown in Fig. 3. Fig. 10(b) mi Profiles for supercritical airfoil at A deg incidence.
Comparisons of the wall stresses are shown in Fig. 4. The Experimental Nakayama: o;SKE: ; MKE:
SKE model degrades in accuracy in the adverse pressure gra- for x/C<1.000, y* = y; for x/C>1.000, y* = y - y c
Re = 1.2x10 6
dient zone as the separation point is approached and fails to
change sign (i.e., no separation). The MKE model departs
from the SKE predictions in the adverse pressure gradient observed, it is not responsible for the important discrepancies
region and agrees much better with the measurements near between the predictions and the experimental data. Similar
separation. The accurate prediction of the location of the conclusions can be drawn for the predictions of k and v,.
separation point already discussed, is reflected in this plot by a Summary. The MKE model provides much better agree-
change in the sign of TW at the correct ^-location. ment with the data than the SKE model for Simpson's
Profiles of the U-velocity comparison is shown in Fig. 5 at problem. There are, however, very significant discrepancies
four stations, two of which are downstream of separation. between results from this model and the data, particularly in
The overall level of agreement is much better with the MKE the separated flow region.
model. The height of the backflow (£/<0) region within the
separation bubble, and the velocity defect within the boundary 3.2 Turbulent Flow Past a Supercritical Airfoil With 4°
layer (y/5<l), are significantly underpredicted by both Angle of Attack at Rec = 1.2 X106
models. Description of the Problem. Nakayama [11] has provided
The turbulent kinetic energy and turbulent viscosities are measurements of flow over the supercritical airfoil shown in
compared in Figs. 6 and 7, respectively. There is some im- Fig. 9, with an angle of attack a = 4°. Measurements were ob-
provement in the k profiles (Fig. 6) predicted by the MKE tained in a wind tunnel with walls located at 0.79C above and
model, compared to SKE, only in the separated flow region, below the airfoil. The inlet velocity was 30 m/s, so that the
but there is generally poor agreement with the measurements Reynolds number based on the chord length C was 1.2x 106,
in this region: the predicted level of k is too small, and the and the turbulence intensity was 0.02 percent. The coordinate
predicted region of high k is confined too close to the wall. system in which the data were reported is also shown in Fig. 9.
The turbulent viscosity (Fig. 7) is greatly overpredicted by the Implementation of the Numerical Model. A 95 x 40 or-
SKE model upstream of the separation (this prevents the flow thogonal grid extended 10 chord lengths upstream and
from separating). It is surprising that the vt profiles predicted downstream of the airfoil. The upper and lower boundaries
downstream of separation are almost identical for the SKE were placed at the location of the wind tunnel wall, and the
and MKE models, even though flow separation has occurred grid conformed to the airfoil surface. Of the 95 control
with the MKE model. volumes in the x, direction, 33 were clustered on both the top
Influence of Grid Refinement of the Predictions. Calcula- and the bottom surfaces of the airfoil. Such a grid was found
tions of the Simpson flow were performed on various grid to provide adequate resolution.
sizes in order to investigate the influence of grid refinement on The inlet velocity and kinetic energy (isotropic turbulence
the accuracy of the predictions. Figure 8 illustrates the velocity assumed) were matched with the experiment. The inlet dissipa-
profiles obtained with MKE at a station located in the recir- tion was selected to make the turbulent viscosity 3 to 4 orders
culating flow region for (60 x 35), (60 x 21) and (95 x 21) grids. of magnitude smaller than its maximum value near the airfoil.
Although a small effect of grid size on the predictions can be Zero normal velocity and slip conditions were applied on the
0.02
O
«-- o.oo
-0.02
-0.04
-0.06
0.0 4.0 S.O 0.0 4.0 8.0 0.0 4.0 8.0 0.0 4.0 8.0 -6.0 -3.0 0.0 3.0 0.0 3.0 0.0 3.0 0.0 3.0 6.0
(vB/U*) x 1000 W £ ) x 1000
Fig. 11(a) vv Profiles for supercritical airfoil at 4 deg incidence. Fig. 12(a) uv Profiles for supercritical airfoil at 4 deg incidence.
Experimental Nakayama: o ; SKE: ; MKE: Experimental Nakayama: o ; SKE: ; MKE:
for x/C < 1.000, y* = y; for x/C>1.000, y* = y-yc for x/C< 1.000, y* = y; for x/C> 1.000, y* = y-yc
Re = 1.2x10 6 Re = 1.2x10 6
0.02
J i 0.00 iio.oo
-0.02
-0.04
-0.06 -6.0 -3.0 0.0 3.0 6.0-3.0 0.0 3.0 0.0 3.0 0.0 3.0
0.0 4.0 8.0 0.0 4.0 B.O 0.0 4.0 8.0 0.0 4.0 8.0
(mi/t/^) x 1000
(CO/D^) x 1000
Fig. 11(b) vv Profiles for supercritical airfoil at 4 deg incidence. Fig. 12(b) uv Profiles for supercritical airfoil at 4 deg incidence.
Experimental Nakayama: o ; SKE: ; MKE: Experimental Nakayama: o;SKE: ;MKE:
forx/C<1.000, y* = y; for x/C>1.000, y* = y-yc forx/C<1.000, y* = y; for x/C>1.000, y' = y-yc
6
Re = 1.2x10 6 Re = 1.2x10
wind tunnel walls. With this boundary condition, the flow far reverse is true in the wake (Fig. 12(b)). The magnitude, and
away from the airfoil is nearly inviscid, but the disturbance of the decay rate behind the airfoil, of the shear stresses is better
the pressure around the airfoil by the wall is accounted for in predicted than the normal stresses.
the model. It would have been too expensive, computational- Summary. For Nakayama's problem, in which both
ly, to resolve the viscous boundary layer growth along the tun- strong streamline curvature and adverse pressure gradients oc-
nel wall. cur, some stresses are better predicted by the MKE model and
Comparisons of turbulent stress measurements and predic- some by the SKE model. The MKE model tends to be better
tions are presented along j'-profiles at several x/C stations be- adjacent to the airfoil and the SKE is better in the wake. While
tween 0.893 and 1.40. x/C< 1.0 denotes stations on the airfoil the general stress levels are reasonably well predicted, the pro-
and, at such stations, the ^-coordinate lies normal to the sur- files are incorrect in the wake.
face of the airfoil (see Fig. 9).
Presentation and Discussion of the Results. Figures 10(a)
and 10(b) compare the measured and predicted uu stress. For 4 Conclusions
stations on the airfoil, the SKE model provides more accurate The performance of the standard k-e model (SKE) as well
predictions, both above (y/C>0) and below the airfoil. Just as that of a modified k — e model (MKE), is evaluated for two
beyond the trailing edge (x/C= 1.02), both models give similar case studies that involve strong adverse pressure gradients
predictions, but further into the wake, the MKE stresses decay and, for one of the cases, strong boundary induced streamline
too quickly. Neither model however provides good agreement. curvature. The purpose of the modification in the MKE model
For example, neither model predicts the double peak of uu in is to sensitize the dissipation equation to the normal stresses.
the wake, that probably results from the advection of uu by For the first problem, involving a sufficiently strong adverse
the streams from above and below__the airfoil. pressure gradient to induce separation, but with no boundary
The levels of the normal stress vv are in better agreement induced curvature, the modified (MKE) model induced
with the measurements (Figs. 11(a), \\(b)), but the double separation while the standard k — e (SKE) model did not. The
peak peak that exist at the trailing edge again disappears failure to predict such an important flow feature represents a
within 1 percent of the chord length. Clearly, the eddy major deficiency of the standard k—e model.
viscosity approach depends too strongly on local effects, with The second problem, of flow over a supercritical airfoil at 4
too little memory of upstream conditions. For this vv stress deg angle of attack, has a strong boundary curvature and in-
component, the MKE model predicts somewhat better levels duces strong adverse pressure gradients on the suction surface.
than the SKE model. In this case, the stresses predicted by the MKE model were
Of particular interest are the uv stresses shown in Figs. slightly more accurate for the flow near the surface, but the
12(a) and 12(b). The MKE model is significantly bet- SKE model provided more accurate stresses in the wake. The
ter than the SKE model on the airfoil (Fig. 12(a)), while the level and wake decay rate of the stresses were roughly correct